CMS report - December 2017

by Scott Wales

Holger Wolff and I at the CMS team have been setting up a high-resolution atmosphere climate configuration for use by researchers both in ARCCSS and the new Centre of Excellence for Climate Extremes. This global model uses a 'n216' grid (432 x 324 grid points, or approximately 1 x 0.5 degrees), double the 'n96' resolution used in the ACCESS 1 models.

Having access to higher resolution models gives researchers a closer view on the processes that drive climate extremes, one of the driving goals for the new Centre.

The configuration that we're setting up is an exact duplicate of the UK Met Office's GA7.1 HighResMIP configuration. HighResMIP is a CMIP6 project looking at the effects of model resolution.

With an exact duplicate of a Met Office run we can perform experiments locally and use existing Met Office results for comparison, without having to spend supercomputer resources on running our own high resolution control runs.

To ensure the configuration we've got at NCI is an exact match to the Met Office run we're both doing an inspection of the output fields with the help of Peter Gibson as well as using a tool called PReVIOuS, developed by João Teixeira from the Met Office's collaboration team. Two model runs performed on different supercomputers are never going to be bitwise identical, so PReVIOuS uses a statistical comparison.

Each of the sites performs a short ensemble run. Each grid point in the inputs are randomly perturbed by 1 unit of least precision, which is the smallest possible difference between two floating point numbers. Running the ensembles then shows the effects of small perturbations, like those caused by running with different compilers. Statistics are then calculated to see if the perturbations at each site are sampled from the same distribution - if they are then we have a good match.

Currently we have two projects that are planning to use the HighResMIP configuration. The development of the CABLE land surface model has identified what is believed to be a bug in the atmosphere-land surface coupling. To better understand the effects of this bug we're planning on running high resolution comparisons between the CABLE and JULES land surface models.

Our second project using the high resolution model is setting up a Transpose-AMIP style configuration, allowing researchers to run hindcasts from an arbitrary start date using the n216 configuration. This setup initialises the model from a reanalysis data set (currently ERA-Interim) and can be used to explore in detail the processes around specific extreme climate events.

More information about these runs, including the Rose job ID required to run them for yourself, is available on the CMS wiki here. At the time of writing these configurations are still being scientifically validated. As always you can email us at climate_help@nci.org.au for assistance with these or any other model used at ARCCSS.

UNSW logo ANU logo Monash logo UMelb logo UTAS logo